… | |
… | |
3 | */ |
3 | */ |
4 | |
4 | |
5 | #include <cstring> |
5 | #include <cstring> |
6 | #include <cstdlib> |
6 | #include <cstdlib> |
7 | |
7 | |
|
|
8 | #include <glib.h> |
|
|
9 | |
8 | #include <tr1/unordered_set> |
10 | #include <tr1/unordered_set> |
9 | |
11 | |
10 | #include "shstr.h" |
12 | #include "shstr.h" |
|
|
13 | #include "util.h" |
11 | |
14 | |
12 | // NOTE: even with lots of stuff loaded, we do not usually have >>20000 strings. |
|
|
13 | // maybe refcounting is just overhead? |
|
|
14 | |
|
|
15 | struct hash |
|
|
16 | { |
|
|
17 | std::size_t operator ()(const char *s) const |
|
|
18 | { |
|
|
19 | unsigned long hash = 0; |
|
|
20 | unsigned int i = 0; |
|
|
21 | |
|
|
22 | /* use the one-at-a-time hash function, which supposedly is |
|
|
23 | * better than the djb2-like one used by perl5.005, but |
|
|
24 | * certainly is better then the bug used here before. |
|
|
25 | * see http://burtleburtle.net/bob/hash/doobs.html |
|
|
26 | */ |
|
|
27 | while (*s) |
|
|
28 | { |
|
|
29 | hash += *s++; |
|
|
30 | hash += hash << 10; |
|
|
31 | hash ^= hash >> 6; |
|
|
32 | } |
|
|
33 | |
|
|
34 | hash += hash << 3; |
|
|
35 | hash ^= hash >> 11; |
|
|
36 | hash += hash << 15; |
|
|
37 | |
|
|
38 | return hash; |
|
|
39 | } |
|
|
40 | }; |
|
|
41 | |
|
|
42 | struct equal |
|
|
43 | { |
|
|
44 | bool operator ()(const char *a, const char *b) const |
|
|
45 | { |
|
|
46 | return !strcmp (a, b); |
|
|
47 | } |
|
|
48 | }; |
|
|
49 | |
|
|
50 | typedef std::tr1::unordered_set<const char *, hash, equal> HT; |
15 | typedef std::tr1::unordered_set<const char *, str_hash, str_equal> HT; |
51 | |
16 | |
52 | static HT ht; |
17 | static HT ht; |
|
|
18 | |
|
|
19 | static const char *makevec (const char *s) |
|
|
20 | { |
|
|
21 | int len = strlen (s); |
|
|
22 | |
|
|
23 | const char *v = (const char *)(2 + (int *)g_slice_alloc (sizeof (int) * 2 + len + 1)); |
|
|
24 | |
|
|
25 | shstr::length (v) = len; |
|
|
26 | shstr::refcnt (v) = 1; |
|
|
27 | |
|
|
28 | memcpy ((char *)v, s, len + 1); |
|
|
29 | |
|
|
30 | return v; |
|
|
31 | } |
|
|
32 | |
|
|
33 | const char *shstr::null = makevec ("<nil>"); |
|
|
34 | |
|
|
35 | // what weird misoptimisation is this again? |
|
|
36 | const shstr undead_name ("undead"); |
53 | |
37 | |
54 | const char * |
38 | const char * |
55 | shstr::find (const char *s) |
39 | shstr::find (const char *s) |
56 | { |
40 | { |
57 | if (!s) |
41 | if (!s) |
… | |
… | |
66 | |
50 | |
67 | const char * |
51 | const char * |
68 | shstr::intern (const char *s) |
52 | shstr::intern (const char *s) |
69 | { |
53 | { |
70 | if (!s) |
54 | if (!s) |
71 | return s; |
55 | return null; |
72 | |
56 | |
73 | if (const char *found = find (s)) |
57 | if (const char *found = find (s)) |
74 | { |
58 | { |
75 | ++refcnt (found); |
59 | ++refcnt (found); |
76 | return found; |
60 | return found; |
77 | } |
61 | } |
78 | |
62 | |
79 | int len = strlen (s); |
63 | s = makevec (s); |
80 | |
|
|
81 | const char *v = (const char *)(2 + (int *)malloc (sizeof (int) * 2 + len + 1)); |
|
|
82 | |
|
|
83 | length (v) = len; |
|
|
84 | refcnt (v) = 1; |
|
|
85 | |
|
|
86 | memcpy ((char *)v, s, len + 1); |
|
|
87 | |
|
|
88 | ht.insert (v); |
64 | ht.insert (s); |
89 | |
|
|
90 | return v; |
65 | return s; |
91 | } |
66 | } |
92 | |
67 | |
93 | // periodically test refcounts == 0 for a few strings |
68 | // periodically test refcounts == 0 for a few strings |
94 | // this is the ONLY thing that erases stuff from ht. keep it that way. |
69 | // this is the ONLY thing that erases stuff from ht. keep it that way. |
95 | void |
70 | void |
… | |
… | |
118 | { |
93 | { |
119 | HT::iterator o = i++; |
94 | HT::iterator o = i++; |
120 | const char *s = *o; |
95 | const char *s = *o; |
121 | ht.erase (o); |
96 | ht.erase (o); |
122 | |
97 | |
|
|
98 | int len = length (s); |
|
|
99 | |
123 | //printf ("GC %4d %3d %d >%s<%d\n", (int)ht.size (), n, shstr::refcnt (s), s, shstr::length (s)); |
100 | //printf ("GC %4d %3d %d >%s<%d\n", (int)ht.size (), n, shstr::refcnt (s), s, shstr::length (s)); |
124 | free (-2 + (int *)s); |
101 | g_slice_free1 (sizeof (int) * 2 + length (s) + 1, -2 + (int *)s); |
125 | } |
102 | } |
126 | else |
103 | else |
127 | ++i; |
104 | ++i; |
128 | } |
105 | } |
129 | |
106 | |